1 AD-A241 357

AGLrARD.PDMI 9OR io FMR AO6E UERH& MEV xb.OP

7 RUE-ANCLE 9220 lEVIJY SUR SEINE FRANCE

-AG DAIORYll wowRI 4

TechnicalEvaluation Report on C-Knowledge Based SystemApplications forGuidance and Control(Application des Syst~mes ' Base de Connaissancesau Guidage-Pilotage)

-¢ NORTH ATLANTIC TREATY ORGANIZATION

ID ac

_ ,._ -_._,,__. Dis O tribution and Availability on Back Cover

AGAW401294

AU~rSORY Ul"~ FOR AESPME RESEARCH & DEVEOPMEN7 RUE ACI 1E 92200 NEU."LY SUR SENE FRICE

AGARD ADVISORY REPORT 284 - - .- tv a.

Technical Evaluation Report onKnowledge Based SystemApplications for ' 3tt'--. f

Guidance and Control(Application des Syst~mes a Base de Connaissances DJie

au Guidage-Pilotage)

Dr Franco CanepaALENIAGruppo Aerei DifesaLaboratorio Intelligenza ArtificialeCorso Marche 4110146 Torino, Italy

The Guidance and Control Panel's 51st Symposium was held at theInstituto Nacional de Industria, Madrid, Spain from 18th to 21st September 19All papers presented at the Symposium were compiled asConference Proceedings CP 474.

_, N . North Atlantic Treaty Organization- Organisation du Trait6 de I'Atlantique Nord

91-13000

The Mission of AGAIR)

AcctooiCb1astif is(ARDisshig enhele cp a d~ isg wdofthNAT Axw nasi eflfddsof ci nec d odb~ isingd o acouh fo d- oon voes:

-Recommwiog effecc m for the I nas; to eC their researen and devdopmew capailties for thecoool heueitofdwhNATOcin

- Proiwsck m f dmiia aiceada ce~io thefiiw'.Comnituc iiited ofacrospacrescard' anderaopmew! (taith Fzicular regd to0t i waycd

- Coinnousl simlauiog advanes in the ar s c sOMnt relerant gosrengthezing the conunoi defence postme;

- Jmpovnig dhe cooperaton among memober nations: in afrospace research and dLopk~mm

- Exchnge cisciemtific and wchn" ainfomauio

- Prmiding assisanc to member natios for the purpose of increasing thewr scientif c and technical pocnitul

-Rmierinescicntiflcandedmnica astancasrequesdtooherNATObodiisand to member nationsin connecion

stith research -. d deveopment problems in the aerospace flekL

The hiebs authority ithin AGAR is the Katonal Delegates Board consisting of officially appointed seaio representativesfrom each member natonThe mission of AGARD is carrid out throughi thePands v~d" arc composed of experts appointedby the National Deleates. the Consultant and ExchanWProgrammec and the AerospaceApplications Studies PromrmmeTheresults of AGARD %wk are reported to the member nations and the NATO Authorities through the AGARD series ofpublications of -Iida this is one-

Participation in AGARD acti~ities is by im~itation only and is normally limited to citizens of the NATO nations.

The content of this publication has been reproduceddirectly from material supplied by AGARD or the authors.

Published July 1991

Copyright C) AGARD 1991All Rights Reserved

ISBN 92-835-0624-3

Printed byv5pecialisedl Printiqngerices Liied40 ChigtvellLane. Loiiglton, Esse-x JGIJ3TZ

Prefakce

Ibeeobnaofi ncs innibat su and task conleziq;in C.- fac ofi-cem mum lI~iCUisa basset the stagefor dcvdopmen o cmit iwsptawu segnion approeches invbng wue of . knowledge based teduwog) The field ofArtifficial IuadiEpce (M) is becoming soldified as a sen and the sednios"ic aspect ae de-aeopi rapiI'mbimpliations for gidance and cootrol we aiornou& Practical applications ofllY am critically dependent on athacedardiietnes compuaff Prcesibg du isanti Wqeraon concepms Recent adw s in digital conmputation techai-piesincludfing dama base mnuqenwjw, represent the core enabling tedmnology necessary fior the devepment oflg bin motwdedesist coCCP5 ', U ultimately lead so maor new niliury capabilities Efficient tactical inforwatio management andeffectise pilot interaction are essential. Pilot deiio aiding. combat automation, sensor Winon and on-board tactical battlenunagemen concepts offr the opportnizy for substantial missi effectiveness: inmprovemcnts. Although real-tune tacticalmilitary applications are relatively few innubr mrlc om d swedevop ntfot:wetxm)Practca miltaryapplicatiom; ofAl wchologyam ofprw)iniyerest. ftced s-m alcV*nanceels alongatihAl limitatoi re consideraLOp -tonalimpfications~.ad critical design srade-offsueaoenlasi-cd.Thissr~nposiumpronided a timely forum for assessing the overall state-of-the-at ofAY applications in The guidance and control area.

A round table discussion to ideozify application issues and opportunities %as held.

Preface

La complexit6 croissante des tfiches mnilitaires; ct des syst~ines d'armes face aux limites inh~rcntes aux capacit~s humaines, apr~par6 le terrain pour le divloppenent d'approches novatrices recourant aux techniques des systimes it base deconnaissances. Le doniaine de l'intelligence artificieflcs'affirinc en tant que science dont les 6l6mcnts technologiques 6volucnttr~s rapidement. Les cons~qucnces pour lcguidagc ct le pilotagesont consid6rables. Les applicafions pratiques de lintelfigenceaftificiclie sont directcment tributaires des ani~liorations des architectures, des techniques du traitement des donndes parordinateurct des concepts d'int6gration. Les progr~s r~cents dans le domaine des echniqups de calcul numcfriques, ycompris; lagestion de bases de donn~cs, repr6sentent le noyau technologique indispensable au d6veloppement de concepts hautementnovatcurs, qui d~boucheront, Ai tcrme, sur de nouvelles et-importantes applications militaires. La gestion efficace desinformiations tactiques et la bonne interaction pilote-syst~me sont essentielles. L'aide 5i ]a d~cision pour le rilote,I'automatisation dui combat, la fusion des capteurs et les concepts de la gestion tactique de la bataille par des moy-.as embarqu&souvrent la voie 5 une amdlioration substantiellc de l'cfficacit6 opdrationnellc. Bien que los applications temps riel soient encorerelativement rares, un certain nombre de projets de ddveloppeznents, tant exploratoires qu'avancds, sont en cours a I'eureactuelle. Les applications militaires des technologies do I'intelligence artificiellesont d'un int&nitprimordial. Les atir6liorationsattendues de I'efflcacitd des moyens militaires ont 6t6 examin~cs conjointemcnt avec iks limitations pr~isibles de I'intclligenceartificielle. Les implications op6rationnelles et les compromis critiques 6tablis lors de la conception des systmes oat fait l'objetd'unc. analyse particulirc. Cc symposium a 6t6 ainsi I'occasion pour faire l'valuation. de I'tat de I'art des applications deI'intelligence artificicile dans le domaine du guidage ct du pilotaige.

Une table rondo s'est tenuc dans le but de faire 6merger los applications potcntielles.

Guidance and Control Panel

Chairmai DrEB.Stca Deputy Chiruu MrS.LeekCorprateVicePresdentPB 256Tedmooev~scssienzBruish Aerospce (Dynamnics) Ltd

The Bocing Compamy PO Box 19P0 Box 3707 six IU WnMail Stop 13-43 SzevnageSeattle,WMA 98124-2207 HeMiSGl 2D.AUnited Sttes United King!domn

TECHNICA~L PROGRAMMAE, CO'MITTRE

Chairmniax Mr James K.Ranage us

Membrs:. DrAndriBcnoit BEMr Bernard Chaillot FRDrHeinzAWinter GEDr Bruno'Mazzetti ITProfessor Pedro Sanz-Aranguez SPProfessorJohn T.Shcpherd UKDr Elihu Zimet us

HOST PAINEL COORDINATOR

Mdr Carlos A. Garriga. LopezSENER Ingenieria y Sistemas SASpace and Defence Divisionc/ Raimundo Fernadcz Vilfaverde, 65Planta 20Madrd

4 Spain

PANEL EXECUTIVE

Commiandant M. Mouhamad, FAF

Mail from Europe: Mail from US and Canada:AGARD-OTAN AGARD-NATOAttn: GCP Executive Attn: GCP Executive7, rueAncelle APO New York 09777F-92200 Neujilly sur SeineFrance

Tel: 33(l) 47 38 57 80Telex: 610176 (F)

Telefax: 33(1) 47 38 5799

ACKNOWVLEDGEMENTS/REMERCIEMENT'S

The Panel wishes to express its thanks to the Spanish National Delegate to AGARD for the invitation to huld this meeting inhis country and for the facilities and personnel which make the meeting possible.

Lc Panel tient a remercier le D616gu6 D:ational de l'Espagne pr~s IAGARD dc son invitation 5I tenir cette r6union dans sonpays et de. ]a misc ii dispositionde pcrscnnel et des installations necessaires.

Contents

Page

Preface/Pr6face iii

Guidance and Control Panel and Programme Committee iv

I Introduttion 1

2. Objectives 1

3. ThePapers 23.1 Session .: RepresentativeApplications 232 Session 2: Design Concepts and Synthesis Techniques 23.3 Session 3: Related Methods and Techniques 33.4 Session 4: Information Processing and System Architecture 33-5 Session 5: 'Mechanization and Integration Issues 3

4. Conclusions 44.1 The Technology Status 44.2 Integration 4

4.2.1 The Integration between Al Technologies 44.2.2 Integration between Al and Conventional Technologies 4

5. Recommendations 5

6. Glossary 6

7. List of Presentations 6

v

1 - Introduction

The 51st Symposium of the Guidance and Control panel on KNOWLEDGE BASED SYSTEMSAPPLICATIONS FOR GUIDVNCE AND CONTROL, was held in Madrid from 18th to 21st September1990.In spite of the prevalence of military aspects the Symposium also dealt with the civilapplications of G&C. The focus was on the problems for the management of IFRoperations.The goal was to achieve an advanced automation of the Air Traffic Management systemsboth in the direction of the Air Traffic Control and on-board systems for navigation anddecision aiding.in the military systems, complexity is becoming so high that it exceeds the possibilityof being directly managed by the human component.G&C problem solving requires complex split-second decision making based on a vast amountof rapidly changing, incomplete and conflicting information in an environment in whichthreat intensity is quickly growing.In these conditions, the psyco-physiological capabilities of human being are inadequatefor the full exploitation of modern systems potential.One way to overcome these limitations is to put into the "man-machine system" thosefunctions which cannot be performed (by the human) with the required speed andefficiency.A.I. technologies, and knowledge processing in particular, have reached a developmentand consistency level which makes it realistic for us to think about systems capable ofsolving the problem.This appears to be possible in two ways.The first one is the capability to manage the complexity arising from the quantity andvariety of information generated by modern sensors technology.The second depends on the fact that it is now possible, in a meaningful measure, tohandle problems which usually require judgement, interpretation and reasoning aboutfacts and circumstances.These kind of topics, quite difficult to represent and treat with algorithmictechniques, can be tackled with KR technologies.

2 - Objectives

One way for describing the contribution of KR technologies to the solution of theproblem is to underline the possibility to segregate the symbolic from the algorithmiccomputation.In the conventional SW, parts of code presiding at control and decision logics, aredeeply interconnected with coding dealing with the execution of procedures andalgorithms.This generates a rapid increase of SW complexity which imposes severe limits to theirdesign.Using A.I. technology, it is possible to extract the former putting them in sections ofSW representing the knowledge needed f6r the problem solving process and leaving thelatter performed by separated algorithmic and procedural sections of conventional SW.The results are not only less complex and more easily maintainable SW, but also a betterperformance in solving problems for which conventional SW found its limits (non-existence of algorithms; algorithms possible, but too difficult to design nd handle;simple problems but requiring unacceptable computing times etc).Use of A.I. technology in military applications dates back to about 1984.Since then several programs in several areas have been started for proving theeffectiveness of this technology.The objective of the symposium was to have an overview on the results achieved up to thepresent day and focus on their meaning as regards the real problem solving capacity inthe G&C field. The key point was the potential of reducing the pilot workload, due toexcess of information, and helping him to make quick decisions in a highly hostileenvironment.A side goal of the symposium was the focusing on special problems forecasted just at thebeginning of AI applications and now more clearly defined owing to the more substantialresults achieved.The symposium organization was well tailored for these purposes, having been dividedinto sessions emphasizing both the practical applications and a broad spectrum of KRTtopics associated with the G&C problems.

Session 1 - Representative Applications.The systems associated with the four papers of the sessiop are consistent applicationswhich problem solving capacity, at different levels of completeness and integration, hasbeen tested on simulators and/or by extended tests by domain experts.Technologies used include the representation of knowledge about mission and missionmanagement (navigation, tactics, interpretation of information patterns aimed atdiagnosis and situation assessment) cooperating expert systems, image interpretation,real-time expert systems, path planning, etc.

Session 2 - Design concepts and systems techniques.Al technologies associated with the paper of this session are: image segmentation, KBfor their interpretation and test of hypotheses developed by models of objectrepresentation, decomposition of planning problems into algorithms and managementfunctions, techniques for the simplification of AI architectures to reduce computationalneeds for real-time performance, methods and techniques of knowledge draining and real-time systems validation.

Session 3 - Related methods and techniques.

3

speed on standard hardware. As an example, a KB generator encodes the interpreted rulesand facts into procedural code speeding up the inferencing process by several orders ofmagnitude, a clever sophisticated simplification indeed.An interesting example of interdisciplinary approaches which can be used for knowledgecollection was given by paper 25. Knowledge about particular systems is not onlyobtainable by human experts but also by observing and describing the system behaviourusing advanced automatic control technologies (as Kalman filters) to deduct generalrules from quantitative information.A less technical but by no means less relevant paper is no. 26 which gave a review ofgeneral methodologies and a *semi-sequential" approach for knowledge draining from humanexperts, selecting domains for applications, and dealing with knowledge engineers andexpert pre-requisites. The topic is strictly connected with the need of developingstandard procedures for knowledge engineering and the KBS life cycle.More technically but about the same topic paper no. 27 dealt with the standardmethodology for testing and validation of real-time expert systems. This problem ariseswhen the prototype phase is finished and the goal is to produce an industrializedsystem. A methodology for testing and validating the system at the end of eachmeaningful phase of the development (initial, feasibility, development, delivery) wasproposed and analyzed in this paper.

3.3 - Session 3 : Related methods and techniquesThe papers of the session are not strictly associated with the G&C problem but they arenevertheless remarkable for showing the progress of specific technologies which can beused in the G&C.The use of fuzzy logic was treated by paper 31 for fuzzy controllers construction.Neural networks were used for the analysis of aircraft test daa (paper no.33) showinghow this technology can be used for the more general problem of extracting informationby data confirmation.Object oriented view of software (paper 34) provides a framework for integrating KBSwith control systems simulations.

3.4 - Session 4 : Information processing and system architectureTwo papers only (no. 41 and no. 42) were presented in this session.Paper no. 41 is associated with the problem of assisting the pilot during the missionbut differing from papers 11, 13, 23, 24 it does not deal with a partially or totallydeveloped prototype but with the study of an architecture enabling everal "expertagents", dedicated to specific problems, to cooperate for the solution of a complexproblem. This paper confirms the combination architecture/technology already establishedby the previous papers. The multi-agents architecture is implemented by the use ofknowledge and metaknowledge representation, cooperating expert systems technology,communications between them, resources management and conflict resolution.Paper no. 42 presented an algorithm for evaluating the optimal homing point for missileguidance.

3.5 - Session 5 : Mechanization and integration issues

Paper no. 51 presented SEAN, a prototype of an ES for navigation aid onboard combataircraft. It addresses a specific function of the navigation problem, providing adviceof available choices while monitoring the inertial navigation system. An advanced methodbased on object oriented programming has been used for collecting and modelingknowledge. The method is based on the structural decomposition of the knowledge objects,on their modeling in taxonomies related to specific functions.Paper no.52 wasn't presented.The state of the art of McDonnel Douglas project ICAAS was presented by paper no. 53.The topic once again is the real-time aid to the pilot of a combat aircraft and thepurpose of the project is to demonstrate the power of AI technology in building systemsable to permit the fighting and survival of tactical fighters when outnumbered by asmuch as four to one by enemy aircraft during air combat engagement. Problems addressedby the system are again focused on the integration of "intelligent systems" with on-board systems, sensors control, data correlation of real-time performance, reasoning toevaluate expected reaction to each tactics and advanced man machine interface. Thispaper presented a different approach from paper 24. In this case the choice has been tofully exploit the potential of advanced AI aspects solving the computational problem notby the semplification of AI structures but developing new powerful hardware (riscarchitecture and with enough MIPS to handle the computational demand). Theaccomplishment level of the system can be better appreciated by the fact that an FIStest bed aircraft is being modified to incorporate the system. Initial flight testingis forecasted for 1991.The main goal of the project described by paper no.54 is the transferring of knowledgeabout flight control systems from the designer to the personnel who has to test andmaintain them. The presented system is a knowledge-based tool able to collect knowledgeused in the system, hardware and software design and prepare intelligent manuals fordiagnosis helping. The system includes model of behaviour and functioning of the systembeing developed.Integration is again the main topic of paper 55; it is applied to inertial and imagesensors for long range missiles guidance. Information gathered by sensors can be usedfor navigation updating and target reconnaissance. KB are used for image interpretationwhile the final extraction of navigation data from th processed and interpreted IR imageinformation, (and their combination with the inertial sensors data) is based onconventional optimization and filtering techniques.

4

4 - Conclusions

4.1 - The technology status.By the stand point of technology status assessment, the objectives of the programcommittee have been fully achieved.The symposium has represented in a meaningful way most of the aspects of Al applied toG&C.A striking-fact has been the evidence of real, effective experience of merging AI anddomain competence.This means that in a brief time span of 5-6 years Al technologies have definitely leftthe AI Lab environment and have been incorporated in the conceptual and pratical"tools" of the active users.The complexity of presented systems and their degree of evolution are the evidence thatAl technologies are now understood, diffused and used with a meaningful uniformitythrough the different environments which produced the presented applications.They can be considered as established -technologies at least as regards how much has beenproduced by Al research.

4.2 - Integration.

A second goal achieved by the congress is the evidence of the progress made not -only bythe capability of transforming the conceptual models generated by AI in real objects,but also that of integrating, in a surprising variety of ways, AI technologiesthemselves.The broad spectrum of topics presented at the congress ranges from highly integratedsystems to KR use for specific problems and offers a good point of view for evaluatingone of the major evolhtions of AI technology, i.e. the integration between itstechnologies and its integration with "conventional" technologies.

4.2.1 - The integration between AI technologies.Complex problems can be treated by KBS able to "use" other KBS, each one dedicated to aspecific task in which the main problem has been decomposed.This is a high-level integration which is followed by the integration of specifictechnologies used in the solution of single tasks associated to each KBS.Examples can be given by the combination of different reasoning (about time, space,causal or fuzzy reasoning, reasoning-about event and actions etc.).Communication structures between different ES (blackboard system) have been developed tomeet the requirements of designed KBS architectures.Further integrations between Al technologies have been caused by the need of creatingeffective interface with KBS.Speech processing (to communicate with the system) and cognitive process representation(for decision making about information to be shown to the pilot according to thecircumstances) are remarkable examples.Interesting perspectives not yet fully explored are arising from the integration betweenKBS and neural networks.

4.2.2 - Integration between AI and conventional technologies.It can be assumed that AI structures not connected with conventional systems are notmeaningful in terms of industrial applications.The industrial exploitation of AI technology is strictly associated with the combinationof both technologies into the so called "hybrid systems".They are systems in which the part difficult to be implemented using conventionaltechnologies is contained in the section based on AI (reasoning processes and decisionmaking) while the conventional sections takes care of the algorithmic processing ofinformation.A typical approach is represented by KBS able to recognize the type of problem, selectmodels and algorithms coherent with the current problem, locate and extract data to beused, feed and activate the models and use the result for continuing the inferenceprocess.The whole process is quite general and well known even if relatively easy to implementonly in reasonably low complexity systems.The integration between AI and conventional technology is not to be intended only assoftware integration but also as an integration between hybrid systems and physicalequipment.Because AI offers the possibility to process the enormous volume of information producedby sensors, it is reasonable to think that sensors themselves may be re-designed for thepurpose of generating more or different information which can be processed by Altechnology now.The congress permits us to draw some conclusion at least about some basic points ofview:• KR technology is established and diffuse enough to be considered a reality.* All the projects pregented suggest that it is worth continuing to use AI and develop

its technologies.• Projects based in AI established technology are not risky anymore. At least they

are not more risky (perhaps even less) than conventional large software projects.Practical limits exist however, and they depend on the still scarce knowledge about

specific technological areas.• No general and/or defined solutions are available on these areas and project risks

can be managed by "ad hoc" solutions only.Research and applications in these areas are worthwhile anyway because the collectedexperience can contribute to reasonably established solution methods.

5 - RecommendationsA first general recommendation arises from the fact that AI can be seen as a "twofoldtechnology".AI, itself is more a set of problem-solving technologies than a branch of the computerscience.The focus of AI is on the domain problems not on the computer science tools, it merelyuses them.considering this aspect it can be worthwhile to diffuse AI problem solving capabilitythrough the G&C world in a more organized and systematic way.The-KG&C Panel WG/l1 on the "Knowledge based G&C functions" can be the convenient toolfor implementing a cross-fertilization-process between these ideas and experiences.More specifically most of the advanced topics require research which can be moreeffectively performed through the common work between groups already active in the G&Cfield.Some advanced topics are:

a) Real time systems.AI applications in domains requiring real time performance create a set of new andunexplored problems different from those KBS which operate on static information, andsolution time is not a critical issue

As examples of problems posed by the real-time needs it is possible to list:

. the need to recover dynamically input data from a variety of sensors* the need'of continuing to operate even if a part of the system fails, (monitoring

systems should operate indefinitely until the intervention of a human operator)• the need to manage asyncronous events

real time systems should accept input data generated by unforseen events, processthem according to their importance even if the current input analysis has to be

interrupted and re-programmedthe need to reason concerning temporal data and therefore have models of knowledgeabout timethe need to concentrate the reasoning on specific events when the time factorbecomes critical

. the consequence may be the need to consult specialized konwledge sources, to modifythe sensor subset currently considered by the system and/or to change the frequencywith which the data are considered

• the need to have, at any moment, the best possible answer.The reasoning system should have the capability to find answers using a trade-offbetween the time constraint and the answer accuracy

b) Neural networkNeural network (NN) potential is now investigated as regards its typical applicativeareas (pattern recognition, solving of combinatorial optimization, data classification).Really interesting is the extension of pattern recognition to the more general meaningof recognition of patterns of information such as data from sensors, informationproduced by simulation models etc.This permits us to think about NN like pre-processors able to assert facts (function ofthe recognition process) to be used by KBSIn other words NN can be used to extract higher level information from low lwvwlinformation even not being able to explain the recognition process.

c) Machine perceptionFor the purpose of G&C problem solving a set of advanced AI topics can be labelled underthis title and, for this paper, they can be loosely divided into machine sensing andman-machine interface.Both share the same fundamental issues, the first being particularly associated withbehaviour which we are used to consider typical of consciousness, and in the second caseperception is aimed at a better interaction with the human being.As regards machine sensing it can be summarized by the capability of perceiving theexternal world from environmental stimulis and its reaction to them.This involves capabilities like collecting information by sensors, interpreting resultsof their processing, and using them to make the more or less autonomous decisionrequired by current circumstances.In some sort of way this kind of process is already evident in the systems presented atthe congress.In the area of situation assessment computer vision could aid in the identification ofthe objects in a scene and the perception of the environment in a given period of timeor space. It can also be used to determine the causal relationship between objects (anexample can be the corresponde5nce between maps and sensed data, range finding, terrainfollowing and path planning).General purpose vision systems are not available and a good deal of research andtechnology should be developed to process visual data, extracting physical data from theimage, labelling significant objects and describing them symbolically.One of the main goals of vision research seems to be how high-level knowledge andinferential procedures contribute to the vision process and how to implement algorithmsthat will allow an interaction between the lower image data and the high-level symbolicknowledge.Man machine interface also involves many perception problems, in two ways particularly.The first one is in making the communication with the computer fast and reliableavoiding the need of stylized and formal languages. This is the area of naturallanguage and speech understanding.

6

The difficulties lie in the fact that understanding what is said depends considerably onthe information and the context which is now available to the computer only in toolimited way for such a purpose.Speech understanding is also difficult because of problems with grammatical errors,accents, noise, separation of words etc.The second one is associated with the fact that the goal of sending to the end user thecorrect information under the correct circumstances, requires some knowledge about themental processes activated by the man.Stated in other words knowledge about why and how man reacts to external stimuli, whatkind of errors he makes and why, how the reasoning changes in stress conditions etc.should be extended. It is essential to know more about what and how to send to the pilotaccording to the circumstances.This is an important part of the knowledge about cognitive processes which needs to bedeeply investigated with the aim of designing and producing sound and reliableinterfaces.

d) Knowledge engineering methodologiesThe current status of software development concerning KBS is reasonably satisfactory asfar as research and the first stages of demonstrators goes.Concerning the further stages of development, from prototypes to operational systems,the production of KBS and systems incorporating KBS still faces many new and unsolvedproblems in order to met industrial quality standards.The final quality of the produced software depends obviously on the quality of knowledgeengineering process, i.e. the way in which the human expertise is collected, expressedand formalized. The process itself is strongly impacted by the choices made in the veryearly stages of development.It is worth, therefore, to promote the growth of standard engineering procedures forKBS.The main topics of the question seem to be:• consistency and completeness of knowledge bases

their validation and maintainabilitytheir portability from one hardware and software environment to anotherthe reduction of knowledge engineering process costs

. the reduction of the bottleneck represented by the high specialized and prizedpersonnel (knowledge 2ngineers).

From the technological point of view, the development of a standard methodology shouldinvolve:* a review of the state of the art on the subject* the identification of development and life cycle model for systems embedding KB

technologiesthe identification of standard development tools and interfacesinvestigate the structure of a knowledge base with emphasis on its consistentmaintenance and efficient use

• investigate how to implement methods for porting and re-using of knowledge basesdevelop standard procedures and methodologies for integration of KBS andconventional tools and systemsinvestigate the methodologies for quality control and maintenance of KBS.

6 - GlossaryAI :Artificial IntelligenceES :Expert SystemIR :InfraredIFR :Instrumental Flight RuleG&C :Guidance and ControlKB :Knowledge BaseKBS :Knowledge Base SystemKR :Knowledge RepresentationKRT :Knowledge Representation TechnologyMIPS :Million Instructions Per SecondNN :Neural NetworkRISC :Reduced Instruction Set Computer

7 - List of presentations

Session 1 - Representative ApplicationsChairman : Dr. H. Winter (GE)(11) The Pilot's Associate exploiting the intelligence advantage. D.I.1OLMES et al. -Lockheed Aeronautical Systems Company, CA, US.(12) SC13 : A knowledge based system for on-board assistance with the interpretation ofinfrared images. D.MORILLON et al. - SAGEM, FR(13) Path generation and evaluation for a mission planning expert system. F.LUISE et al.- Aeritalia, Defense Aircraft Group, Torino, IT.(14) Knowledge based Cockpit Assistant for IFR Operations. R.ONKEN - Universitat derBundeswehr, Minchen, GE.

Session 2 - Design Concepts and Synthesis TechniquesChairman Mr. B. CHAILLOT (FR)(21) An automatic target acquisition system. P.VALERY - Thomson-CSF (FR).(22) Constraint management requiremerts for on-line aircraft route planning. U.TEEGEN -DLR, Institut fdr Flugfdhrung, Braunschweig, GE.

kY

(23) Planning and planning management for autononcz and semi-a~o= vebicles.M.B.ADPA'S et al- - The Charles Stark DrprLaboratory-(24) Intelligence real tim knowledge based imf!iS~t =issico =anagewdent- G.F.MZISS -Boeing Military Airplanes, WA, O3S.(25) Knowledge acquisition for expert syst-em =zing statistical methofs. B.L.B=-ZIN Ctal. - AT & T Bell Laboratories, Xiddletovn, CS.(26) Knowledge extraction methods for the &development of expert xrsteax-s N-.PEOZZ et al.- Directorate of Aerospace Studies, Air F'orce Systems Ckvmmad, MO, CS-(27) A methodology for producing validated real tine en.ert zy.. ens. S..WS et al- -Foyal Aerospace Establishment, Farmboroogh, C1.

Session 3 - Related 15ethods and TechniquesChairman :Dr. A. BENOIT (BE)(3nstrevio Niofa doe aecpicts eospdeignin fuazzyd cotoles .21E. ta(31) Aut reviwa of soonicaspespail on dsiging SuP.otoles .P.L~ ta(32) Paper 32 Withdrawn.(33) A neural network for the analysis of aircraft test data. J-3-GIE III et al - theUp'-.ersity of Tennessee Space Institute. Ai Lab., 7W2, CS.(34) An ADA frambework for the integration of NBS and control systen simulations.M.J.CORBIN et al. - Royal Aerospace Establishment, FaruAboro= , . l.

Session 4 - Information Processing and Syvsten ArchitectureChairman :Professor J.T. S~r--2E=RD (US)(41) The murlti-agents' problenatic in the electronic co-pilot. G-GILES et al. - Avions;Marcel Dassault, Saint Cloud, FR.(42) Evaluation of the optimal honing point for missile guidance. B.NR)DLLINI et al.-Fiar, Milan, IT.-(43) Design and simulation of an advanced airborne early warning systAm. C.-.iIUANG etal. - Corporate Research Center, Gr:.an, !rY, US.

session 5 - mechanization and Integration IssuesChairman :Dr. P. SANZ )RANGUEZ (SP)(51) SEAN: a navigation aid expert system for coubat aircraft.D.MORILLON et al. - SAGEH, FR(52) Paper 52 withdrawn.(53) Integrated control and avionics for air s-uperiority: A knowledge based decisionaiding system. D.J.HALSKY et al. - McDonne! Aircraft, St Louis, MO, US.(543% A knowledge based system design/information tool for aircraft flight controlc.ystems. D.A.MACKPLL et al. - NASA, Anes Research Center, EDwards, CA, US.(55) A study of integrated image and inertial sensor sysLea. R:KOCH et al.-Bodenseewerk Geratetechnik GmbH, fiberlingen, GE.

I GARD-AR-2840 ISBN 92-S35-0624- 3CLASSIFID

Nx'"L A A wv&Trfrwe ,nizzIF 7 -Ancdfi 922W Nc ,v, Sei=e, FrYac

'MC] T NICAL EVALUATION REPORT ON KNO\ EDGE BASEDSYSTEMI APPLICATONS FOR GUIDANCE AND CONTROL

7..Pe

D.- F.-o Cn July 1991

" 1..thes-E&sAddress j11 PagesALENIA. Grppo Aeri DifesaLaboratoio Intelligenza Arfificiale I1 14Corso Marche 41.10146 Torino, Italy

12. DLsIi-ibuion Statenit 'This document is distributed in accordance iith AGARDpolicies and regulations, which are outlined on the

1back coirs of all AGARD publications.!13. Kemords/Descriptors

Al Pilot decision aidingCombat automation Sensor fusionOn-board tactical battle management Tactical information management

14. Abstract

Technical Evaluation Report on the Guidance and Control Panel's 51st Symposium held at the

Instituto Nacional de Industria, Madrid, Spain from 18th to 21st September 1990.

In all, 21 papers were presented including the Keynote Address covering the following headings:

Representative applications;Design concepts and synthesis techniques;Related methods and techniques;Information processing and system architecture;Mechanization and integration issues.

All papers presented at the Symposium were compiled as Conference Proceedings CP 474.

25_____________________

VC______________________________

a .~ . -~ .

<~ oo -n >

zkr

ZP 5E - - 0-

- zn

z 0~ >0 >IL)~ 0

- ~ COO -

-o C z

0 0 .Jo

0 o' C.1 0 0 w7;a a_--

a. C

* -. 0 -o 0

0

.5 o0--.

W20 0 --

00 0j~ 0) o 0N >. Z Ca.-

0 C0 .3&0 z ~ E.0 C 0ro Sz- oo <. -< 0." .-

> C3 -n Le ,t >-ouj

rS z) C)

P2- >. <O o). >O C U

L - W~ r- 0 1 M

I..

Co1

Li c

:3 LO)

cr 0

0: Z = L.

CJ ~

0.0 -i 0

=i - Z 0 =

co .4 - =- .-

0. iZ 10 02 0 U

-c .5 -m o20~~r.~Li 14LiCJ~i-. ~ -